The Recovery Manager has a "Rollback" method with which transactions are aborted. When a transaction aborts, all references of the transaction are removed from the data pages and an abort log record and CLRs are written to the appropriate log file.
During normal operation, the Recovery Manager provides methods to log update and commit operations to the log file. That is, the manager takes care of the format of the update and commit log records.
The real work of the Recovery Manager is after a crash. The "Restart" rooutines take care of the three phases : Analysis, Redo and Undo, and restore the system to a consistent state.
The Recovery Manager also takes care of checkpointing. During checkpointing the dirty page table and transaction tables are written to the log file. After a crash, the analysis phase starts from the log record of the last checkpoint.